Continuity research: two days of experiments and evidence
September 13–14, 2026 · Fourteen radar groups, fifteen engineering topics, thirteen PRs and three pinned commits.
| Local experiment | Result | Scope |
|---|---|---|
| Five dispatcher cases | Base: 3 fail, 2 pass; fix: 5 pass | Real subprocess, fixture CLI |
| Five SDK cases | Old clear: 4 fail, 1 pass; candidate: 5 pass | Real SQLite, remote AsyncMock |
| Seven Paperclip inputs | Two ordinary resumes now select the current full brief | Original selector, normalized-input double; no objective DB |
| Quota and helper policies | 11 pass, including 7 quota-policy tests | Unit tests; no live quota or savings measurement |
Reproduction package · Source audit · Experiment notes · Leads awaiting validation
Eleven product-investigation leads await validation. They are not established development findings or implementation tasks.
How Did an Opening Remark Become the Deliverable? · Why Can an Empty History Still Reference an Old Conversation?